SCRAMBLING AND THE NATURE OF MOVEMENT*
Yuji Takano
Kinjo Gakuin University
1.
Introduction
This paper is concerned with the nature of scrambling and its status in the general
theory of movement. It is well known that (Japanese) scrambling has properties that are
apparently not shared by other kinds of movement. There have thus been a lot of work on
this topic and many proposals have been made to account for those properties (see, among
others, Bošković 2004, Bošković and Takahashi 1998, Fukui 1986, 1993, Grewendorf and
Sabel 1999, Kawamura 2004, Kitahara 2002, Ko 2007, Kuroda 1988, Miyagawa 1997,
2005, 2006, Nemoto 1993, Nishigauchi 2002, Oka 1989, Saito 1985, 1989, 1992, 2003,
2005, Saito and Fukui 1998, Sauerland 1999, Tada 1990, 1993, Takano 1995, 1998, and
Yamashita 2006). Given this backdrop, this article aims to make a contribution toward a
better understanding of the nature of scrambling from a novel perspective, by looking at its
interaction with control in Japanese. I will show that a close examination of scrambling out
of an obligatory control clause (i.e., the complement clause of an obligatory control
construction) in Japanese reveals interesting asymmetries in binding effects that have
previously been unnoticed. I will propose that these newly discovered facts can be
accounted for only if the following two claims hold: (i) unlike the previous view to the
contrary, scrambling out of an obligatory control clause behaves exactly like scrambling out
of a finite clause and (ii) obligatory control involves movement of the controller. I will also
discuss new issues that arise from this proposal about the nature of movement.
This paper is organized as follows. In section 2, I will present previously unnoticed
facts about binding effects with scrambling out of a control clause in Japanese. In section 3,
I will propose an analysis of those facts that relies crucially on the two claims mentioned
above. In section 4, I will discuss issues that arise from this proposal. Finally, section 5
summarizes the discussion.
*
For their helpful comments, questions, and discussion, I would like to thank Jun Abe, Hiroshi
Aoyagi, Tomohiro Fujii, Koji Hoshi, Daisuke Inagaki, Aiko Ishikawa, Howard Lasnik, Roger
Martin, Mikinari Matsuoka, Shigeru Miyagawa, Takashi Munakata, Masashi Nomura, Masao Ochi,
Toshifusa Oka, Kaori Okamoto, Norvin Richards, Mamoru Saito, Koji Sugisaki, Daiko Takahashi,
Kensuke Takita, Hanae Terashima, Hiroyuki Ura, Juan Uriagereka, Akira Watanabe, and Hideaki
Yamashita, as well as audiences at the symposium of the Twenty-Fifth Conference of the English
Linguistic Society of Japan (Nagoya University), Tohoku Gakuin University, Yokohana National
University, and Nanzan University.
Nanzan Linguistics 5, 75-104
©2009 Yuji Takano
Nanzan Linguistics 5: Research Results and Activities 2008 ~ 2009
2.
The Puzzle: Scrambling out of a Control Clause in Japanese
It is well known that there are asymmetries between clause-internal and long-distance
scrambling in Japanese (Saito 1992, Tada 1990, 1993; see also Mahajan 1990 for the same
facts in Hindi). The following examples show that a pronominal element contained in the
subject cannot be bound by a quantificational phrase (QP) in the object:1
(1)
a. * Sokoi-no sotugyoosei-ga mittu-izyoo-no
daigakui-ni
syutugansita.
it-GEN graduate-NOM three-or.more-GEN university-DAT applied
‘Their graduates applied to three or more universities.’
b. * Sokoi-no syain-ga
mittu-izyoo-no
kaisyai-o
tyoosasita.
it- GEN employee-NOM three-or.more-GEN company-ACC investigated
‘Their employees investigated three or more companies.’
Thus, the example in (1a) cannot be interpreted as “there are three or more x, x a university,
such that someone who graduated from x applied to x.” Similarly, the example in (1b)
cannot receive the interpretation “there are three or more x, x a company, such that an
employee of x investigated x.” These are typical cases of weak crossover effects in
Japanese.2
In contrast, the intended bound variable interpretation becomes possible when the
object QP scrambles to the front of the sentence:
(2)
a.
Mittu-izyoo-no
daigakui-ni
sokoi-no sotugyoosei-ga syutugansita.
three-or.more-GEN university-DAT it-GEN graduate-NOM applied
b.
Mittu-izyoo-no
kaisyai-o
sokoi-no syain-ga
tyoosasita.
three-or.more-GEN company-ACC it-GEN employee-NOM investigated
1
A few words about Japanese examples are in order. First, in this article I use soko as a pronominal
element to be bound by a QP. Soko literally means “that place” but I gloss it as “it” for ease of
exposition. Second, following Hoji (2003), I avoid using QPs like daremo ‘everyone’ and subete
‘all’ that can be used to refer to a specific group of entities. Hoji points out that use of such QPs
obscures judgment on bound variable interpretation in Japanese.
2
If the QP is the subject and the pronominal is contained in the object, the QP can bind the
pronominal:
(i)
Mittu-izyoo-no
daigakui-ga
sokoi-no sotugyoosei-o saiyoosita.
three-or.more-GEN university-NOM it- GEN graduate-ACC employed
‘Three or more universities employed their graduates.’
This example permits a bound variable interpretation for soko, so that it can be interpreted as “there
are three or more x, x a university, such that x employed someone who graduated from x.”
-76-
Scrambling and the Nature of Movement (Y. Takano)
A standard approach to the facts in (1) and (2) is to appeal to a necessary condition on
pronominal variable binding to the effect that a pronominal needs to be c-commanded by a
QP if the former is to be bound by the latter. Given this condition, the contrast between (1)
and (2) follows since in (1) the object QP does not c-command the pronominal, whereas in
(2) it does, because of scrambling. In this way, clause-internal scrambling has the effect of
making variable binding possible.
Long-distance scrambling (i.e., scrambling out of a clause) does not show the same
effects. The examples in (3), without scrambling, do not permit a bound variable
interpretation, as expected.
(3)
a. * Sokoi-no sotugyoosei-ga Aya-ni
[Ken-ga
mittu-izyoo-no
it-GEN graduate-NOM Aya-DAT Ken-NOM three-or.more-GEN
daigakui-ni
syutugansita to] itta.
university-DAT applied
that told
‘Their graduates told Aya that Ken applied to three or more universities.’
b. * Sokoi-no syain-ga
Aya-ni
[Ken-ga
it-GEN
employee-NOM Aya-DAT Ken-NOM
kaisyai-o
tyoosasita to] itta.
company-ACC investigated that told
mittu-izyoo-no
three-or.more-GEN
‘Their employees told Aya that Ken investigated three or more companies.’
What is surprising is the fact that the bound variable interpretation does not become
possible even if the object QP of the embedded clause scrambles to the front of the matrix
clause, as shown in (4).
(4)
a. * Mittu-izyoo-no
daigakui-ni
sokoi-no sotugyoosei-ga Aya-ni
three-or.more-GEN university-DAT it-GEN
graduate-NOM Aya-DAT
[Ken-ga
syutugansita to] itta.
Ken-NOM applied
that told
b. * Mittu-izyoo-no
kaisyai-o
sokoi-no syain-ga
Aya-ni
three-or.more-GEN company-ACC it-GEN graduate-NOM Aya-DAT
[Ken-ga
tyoosasita to] itta.
Ken-NOM investigated that told
The same pattern can be seen when the pronominal is contained in the indirect object,
instead of the subject, of the matrix clause:
-77-
Nanzan Linguistics 5: Research Results and Activities 2008 ~ 2009
(5)
a. *Aya-ga
sokoi-no sotugyoosei-ni [Ken-ga mittu-izyoo-no
Aya-NOM it-GEN
graduate-DAT
Ken-NOM three-or.more-GEN
daigakui-ni
syutugansita to] itta.
university-DAT applied
that told
‘Aya told their graduates that Ken applied to three or more universities.’
b. * Aya-ga sokoi-no syain-ni
[Ken-ga
mittu-izyoo-no
Aya-NOM it-GEN employee-DAT Ken-NOM three-or.more-GEN
kaisyai-o
tyoosasita to] itta.
company-ACC investigated that told
‘Aya told their employees that Ken investigated three or more companies.’
(6)
a. * Mittu-izyoo-no
daigakui-ni
Aya-ga sokoi-no sotugyoosei-ni
three-or.more-GEN university-DAT Aya-NOM it-GEN graduate-DAT
[Ken-ga
syutugansita to] itta.
Ken-NOM applied
that told
b. * Mittu-izyoo-no
kaisya-io
Aya-ga sokoi-no syain-ni
three-or.more-GEN company-ACC Aya-NOM it-GEN
employee-DAT
[Ken-ga
tyoosasita to] itta.
Ken-NOM investigated that told
These facts thus indicate clearly that long-distance scrambling does not produce new
binding relations, in sharp contrast to clause-internal scrambling.
The examples in (3)-(6) above have finite clauses as their embedded clauses.
However, Mahajan (1989) pointed out that in Hindi scrambling out of an infinitival clause
exhibits a different pattern from scrambling out of a finite clause. On the basis of Mahajan’s
work on Hindi, Nemoto (1993) closely examines scrambling in obligatory control
constructions in Japanese and concludes that the same holds in this language. Let us
compare (7) and (8) below.
(7)
a. * Sokoi-no sotugyoosei-ga [mittu-izyoo-no
daigakui-ni
it-GEN graduate-NOM
three-or.more-GEN university-DAT
syutugansi-yoo to] sita.
apply-will
that did
‘Their graduates tried to apply to three or more universities.’
-78-
Scrambling and the Nature of Movement (Y. Takano)
b. * Sokoi-no syain-ga
[mittu-izyoo-no
kaisyai-o
it-GEN employee-NOM three-or.more-GEN company-ACC
tyoosasi-yoo
to] sita.
investigate-will that did
‘Their employees tried to apply to three or more companies.’
(8)
a.
Mittu-izyoo-no
daigakui-ni
sokoi-no sotugyoosei-ga
three-or.more-GEN university-DAT it-GEN
graduate-NOM
[syutugansi-yoo to] sita.
apply-will
that did
b.
Mittu-izyoo-no
kaisyai-o
sokoi-no syain-ga
three-or.more-GEN company-ACC it-GEN employee-NOM
[tyoosasi-yoo
to] sita.
investigate-will that did
The examples in (7) are subject control constructions. As expected, the pronominal
contained in the matrix subject cannot be bound by the embedded object QP. In contrast, the
intended variable binding becomes possible when the object QP scrambles to the front of
the matrix clause, as shown in (8). The contrast between (4) and (6) on the one hand and (8)
on the other shows an asymmetry between the two types of long-distance scrambling:
whereas scrambling out of a finite clause does not make variable binding possible,
scrambling out of a control clause does, as Nemoto (1993) observes.
The same effects can be seen with object control constructions. Compare (9) with (10)
and (11).
(9)
a. * Ken-ga sokoi-no sotugyoosei-ni [mittu-izyoo-no
daigakui-ni
Ken-NOM it-GEN
graduate-DAT
three-or.more-GEN university-DAT
3
syutugansuru yoo(ni)] susumeta.
apply
C
recommended
‘Ken recommended their graduates to apply to three or more universities.’
b. * Ken-ga sokoi-no
Ken-NOM it-GEN
tyoosasuru yoo(ni)]
investigate C
syain-ni
[mittu-izyoo-no
kaisyai-o
employee-DAT three-or.more-GEN company-ACC
iraisita.
asked
‘Ken asked their employees to investigate three or more companies.’
3
Following Uchibori (2000), I assume that yoo(ni) appearing at the end of the embedded clause of
the object control construction is a complementizer. See Uchibori 2000 for detailed discussion and
arguments in favor of this position.
-79-
Nanzan Linguistics 5: Research Results and Activities 2008 ~ 2009
(10)
a. ? Mittu-izyoo-no
daigakui-ni
Ken-ga sokoi-no sotugyoosei-ni
three-or.more-GEN university-DAT Ken-NOM it-GEN
graduate-DAT
[syutugansuru yoo(ni)] susumeta.
apply
C
recommended
b. ? Mittu-izyoo-no
kaisyai-o
Ken-ga sokoi-no syain-ni
three-or.more-GEN company-ACC Ken-NOM it-GEN
employee-DAT
[tyoosasuru yoo(ni)] iraisita.
investigate C
asked
(11)
a. ?Ken-ga
mittu-izyoo-no
daigakui-ni
sokoi-no sotugyoosei-ni
Ken-NOM three-or.more-GEN university-DAT it-GEN graduate-DAT
[syutugansuru yoo(ni)] susumeta.
apply
C
recommended
b. ? Ken-ga mittu-izyoo-no
kaisyai-o
sokoi-no syain-ni
Ken-NOM three-or.more-GEN company-ACC it-GEN
employee-DAT
[tyoosasuru yoo(ni)] iraisita.
investigate C
asked
The examples in (9) are object control constructions without scrambling and those in (10)
and (11) are their variants with scrambling of the embedded object to the matrix clause, the
difference between (10) and (11) lying in the landing site of scrambling. The bound variable
reading is impossible in (9) but it is possible in (10) and (11).4 This is another indication
that scrambling out of a control clause behaves differently from scrambling out of a finite
clause.
These observations naturally lead to the generalization in (12).
(12)
Scrambling out of a control clause patterns with clause-internal scrambling.
In fact, Nemoto (1993) tries to derive this generalization from the properties of control
constructions and movement.
However, on closer inspection, we see that the situation is more complicated. Let us
consider the cases in (13) and (14).
4
Nemoto (1993) judges examples like (10) and (11) to be fully acceptable on the bound variable
reading. Although I find (10) and (11) slightly worse than (8), the important point is that (10) and
(11) are much better than (4) and (6).
-80-
Scrambling and the Nature of Movement (Y. Takano)
(13)
a. * Sokoi-no sotugyoosei-ga Ken-ni [mittu-izyoo-no
daigakui-ni
it-GEN
graduate-NOM Ken-DAT three-or.more-GEN university-DAT
syutugansuru yoo(ni)] susumeta.
apply
C
recommended
‘Their graduates recommended Ken to apply to three or more universities.’
b. *Sokoi-no syain-ga
Ken-ni [mittu-izyoo-no
kaisyai-o
it-GEN
employee-NOM Ken-DAT three-or.more-GEN company-ACC
tyoosasuru yoo(ni)] iraisita.
investigate C
asked
‘Their employees asked Ken to investigate three or more companies.’
(14)
a. ?*Mittu-izyoo-no
daigakui-ni
sokoi-no sotugyoosei-ga Ken-ni
three-or.more-GEN university-DAT it-GEN graduate-NOM Ken-DAT
[syutugansuru yoo(ni)] susumeta.
apply
C
recommended
b.?* Mittu-izyoo-no
kaisyai-o
sokoi-no syain-ga
Ken-ni
three-or.more-GEN company-ACC it-GEN
employee-NOM Ken-DAT
[tyoosasuru yoo(ni)] iraisita.
investigate C
asked
The cases in (14a, b) are scrambling variants of those in (13a, b), respectively. It is not
surprising that the latter do not allow a bound variable interpretation. What is striking is that
long-distance scrambling does not make the bound variable reading possible in (14), in
contrast to what we saw in (8)/(10)/(11).5 Given that the cases in (14), just like those in
(8)/(10)/(11), involve scrambling out of a control clause, the contrast between them shows
that the generalization in (12) is not correct.
The point can be strengthened by considering cases where the pronominal is contained
in an adjunct belonging to the matrix clause.6 First of all, as in the case of (1) and (2),
where the pronominal is contained in the subject, clause-internal scrambling of the object
5
As far as I know, the contrast between cases like (8)/(10)/(11) on the one hand and those like (14)
on the other has been unnoticed in the literature. Uchibori (2000) reports a judgment according to
which there is no such contrast (but Uchibori uses daremo ‘everyone’ as a QP; see note 1). I
consulted fourteen speakers (all linguists) and ten of them agreed with my judgment. Three of them
agreed that there is a contrast in the direction indicated here but did not find the bound variable
interpretation in (14) to be as bad as I do. The remaining one speaker found the bound variable
interpretation to be impossible in all cases in (8), (10), (11), and (14). In any case, what is crucial is
the fact that those speakers who detect a contrast between (8)/(10)/(11) and (14) all find (14) to be
worse than the others, not the other way around. This is an important fact that calls for an account.
6
Thanks to Daiko Takahashi for bringing the relevance of such cases to my attention.
-81-
Nanzan Linguistics 5: Research Results and Activities 2008 ~ 2009
QP can make variable binding possible for a pronominal contained in an adjunct. The
examples in (16) below are scrambling variants of those in (15).
(15)
a. * Ken-ga sokoi-no sotugyoosei-no mae-de
Ken-NOM it-GEN
graduate-GEN front-at
mittu-izyoo-no
daigakui-ni
denwasita.
three-or.more-GEN university-DAT called
‘Ken called three or more universities in the presence of their graduates.’
b. * Ken-ga sokoi-no syain-no
mae-de
Ken-NOM it-GEN
employee-GEN front-at
mittu-izyoo-no
kaisyai-o
hihansita.
three-or.more-GEN company-ACC criticized
‘Ken criticized three or more companies in the presence of their employees.’
(16)
a.
Mittu-izyoo-no
daigakui-ni
Ken-ga sokoi-no
three-or.more-GEN university-DAT Ken-NOM it-GEN
sotugyoosei-no mae-de denwasita.
graduate-GEN front-at called
b.
Mittu-izyoo-no
kaisyai-o
Ken-ga sokoi-no
three-or.more-GEN company-ACC Ken-NOM it-GEN
syain-no
mae-de hihansita.
employee-GEN front-at criticized
Consider now cases involving control. The examples in (17) have a pronominal
contained in an adjunct of the matrix clause and a QP object in the embedded control
clause. They do not permit a bound variable interpretation for the pronominal.
(17)
a. * Ken-ga sokoi-no sotugyoosei-no mae-de Yumi-ni
Ken-NOM it-GEN graduate-GEN front-at Yumi-DAT
[mittu-izyoo-no
daigakui-ni
syutugansuru yoo(ni)] susumeta.
three-or.more-GEN university-DAT apply
C
recommended
‘Ken recommended Yumi in the presence of their graduates to apply to three or
more universities.’
-82-
Scrambling and the Nature of Movement (Y. Takano)
b. *Ken-ga sokoi-no
Ken-NOM it-GEN
[mittu-izyoo-no
three-or.more-GEN
syain-no
employee-GEN
kaisyai-o
company-ACC
mae-de Yumi-ni
front-at Yumi-DAT
tyoosasuru yoo(ni)] iraisita.
investigate C
asked
‘Ken asked Yumi in the presence of their employees to investigate three or more
companies.’
When the object QP of the control clause scrambles to the matrix clause, the sentences in
(18) result. They all disallow the intended bound variable interpretation, just like the
examples in (14).
(18)
a. ?*Mittu-izyoo-no
daigakui-ni
Ken-ga sokoi-no
three-or.more-GEN university-DAT Ken-NOM it-GEN
sotugyoosei-no mae-de Yumi-ni
[syutugansuru yoo(ni)] susumeta.
graduate-GEN
front-at Yumi-DAT apply
C
recommended
b. ?*Ken-ga
mittu-izyoo-no
daigakui-ni
sokoi-no
Ken-NOM three-or.more-GEN university-DAT it-GEN
sotugyoosei-no mae-de Yumi-ni
[syutugansuru yoo(ni)] susumeta.
graduate-GEN front-at Yumi- DAT apply
C
recommended
c.? *Mittu-izyoo-no
kaisyai-o
Ken-ga sokoi-no
three-or.more-GEN company-ACC Ken-NOM it-GEN
syain-no
mae-de Yumi-ni
[tyoosasuru yoo(ni)] iraisita.
employee-GEN front-at Yumi-DAT investigate C
asked
d.?*Ken-ga
mittu-izyoo-no
kaisyai-o
sokoi-no
Ken-NOM three-or.more-GEN company-ACC it-GEN
syain-no
mae-de Yumi-ni
[tyoosasuru yoo(ni)] iraisita.
employee-GEN front-at Yumi-DAT investigate C
asked
The ill-formed status of the examples in (18) is unexpected under the generalization in (12).
Thus, the contrast between (8)/(10)/(11) on the one hand and (14)/(18) on the other
undermines the generalization in (12) and requires a different account.
Notice that the presence of an obligatory control structure plays an essential role in
making cases like (8)/(10)/(11) grammatical. If the embedded clause is finite and has a
phonetically null subject coreferential with a matrix element, long-distance scrambling does
not make a bound variable interpretation possible, as shown in (19).7
7
Some speakers seem to find (19) to be slightly better than (14)/(18) but worse than (8)/(10)/(11).
I thank Jun Abe, Daiko Takahashi, and one reviewer for pointing this out to me.
-83-
Nanzan Linguistics 5: Research Results and Activities 2008 ~ 2009
(19)
a. ?*Mittu-izyoo-no
three-or.more-GEN
[proj syutugansita
applied
daigakui-ni
sokoi-no sotugyooseij-ga Ken-ni
university-DAT it-GEN
graduate-NOM Ken-DAT
to] itta.
that said
‘Their graduates told Ken that they applied to three or more universities.’
b. ?* Mittu-izyoo-no
three-or.more-GEN
[proj tyoosasita
investigated
kaisyai-o
sokoi-no syainj-ga
Ken-ni
company-ACC it-GEN
employee-NOM Ken-DAT
to] itta.
that said
‘Their employees told Ken that they investigated three or more companies.’
What about the control structure makes (8)/(10)/(11) grammatical but not (14)/(18)? A
close examination of the relevant examples reveals that the crucial factor distinguishing
grammatical (8)/(10)/(11) from ungrammatical (14)/(18) seems to be the fact that the
pronominal soko is contained in the controller in the former but not in the latter. The correct
generalization thus seems to be (20).
(20)
Scrambling out of a control clause makes variable binding possible only if the
pronominal is contained in the controller.
Why does this generalization hold? In the next section, I will propose that (20) follows from
an interaction of scrambling and movement of the controller under a movement theory of
control.
3.
Solving the Puzzle
I propose that the generalization in (20) can be derived if the following claims hold:
(21)
a. Scrambling out of a control clause patterns with scrambling out of a finite clause.
b. Obligatory control is derived by movement of the controller.
c. The relevant variable binding in (8)/(10)/(11) is licensed by clause-internal
scrambling.
(21a) is an alternative to the generalization in (12) above. (21b) is a movement theory of
control proposed (in different forms) by Bowers (1973, 2008), Hornstein (1998, 1999), and
O’Neil (1995), and argued for in Boeckx (2000), Boeckx and Hornsterin (2003, 2004,
2006), Fujii (2006), and Hornstein (2001, 2003). (21c) is a consequence of (21a) and (21b).
To see how this proposal works, let us consider the derivation of the examples in (8)
shown in (22), where material surrounded by angled brackets indicates copies without
phonetic realization.
-84-
Scrambling and the Nature of Movement (Y. Takano)
(22)
(II) movement of controller
Y-DAT/ACC X-NOM [<Y> <X> <Y> V]
(III) scrambling
V
(I) scrambling
The first important step of the derivation is scrambling of the embedded object Y within the
control clause. This scrambling puts Y in a position c-commanding the subject X of the
embedded clause. Under the movement theory of control, the controller originates from the
subject of the embedded clause and moves to the matrix clause. Given this, the next step is
movement of X to the matrix clause, in accordance with the movement theory of control.
Finally, Y scrambles to the matrix clause.
Given (21a), the second scrambling (i.e., step (III)) has no effects on binding. On the
other hand, the first scrambling (step (I)) is clause-internal scrambling and can affect
binding. Therefore, step (I) of the derivation ensures that Y can bind the pronominal
contained in X.
The object control cases receive a similar analysis. Let us consider (23), which is a
derivation for the examples in (10).
(23)
(II) movement of controller
Y-DAT/ACC Z-NOM X-DAT
[<Y > <X> <Y> V]
(III) scrambling
V
(I) scrambling
Here too, scrambling of Y within the control clause makes the relevant binding possible
under the movement theory of control and further scrambling does not play any role with
respect to binding.8
What is crucial in both (22) and (23) for the pronominal contained in X to be bound by
Y is that long-distance scrambling is composed of shorter scramblings and that an
intermediate scrambling within an embedded clause can produce new binding relations.
That an intermediate scrambling within an embedded clause can produce new binding
relations can be independently seen in cases like the following:
8
In (22) and (23) Y moves past X and X moves past Y. One might think that this situation raises
problems with minimality. However, it is well known that Japanese scrambling does not induce
minimality effects (see, for example, Saito and Fukui 1998, Takano 1995, and Yamashita 2006 for
discussion and specific proposals). Since the issue is not the main focus of this article, I simply
assume this property of scrambling and do not attempt to explain it. See also note 15.
-85-
Nanzan Linguistics 5: Research Results and Activities 2008 ~ 2009
(24)
a. * Ken-ga Aya-ni
[sokoi-no sotugyoosei-ga
Ken-NOM Aya-DAT it-GEN
graduate-NOM
mittu-izyoo-no
daigakui-ni
syutugansita to] itta.
three-or.more-GEN university-DAT applied
that told
‘Ken told Aya that their graduates applied to three or more universities.’
b. * Ken-ga Aya-ni [sokoi-no syain-ga
Ken-NOM Aya-DAT it-GEN
employee-NOM
mittu-izyoo-no
kaisyai-o
tyoosasita
to] itta.
three-or.more-GEN company-ACC investigated that told
‘Ken told Aya that their employees investigated three or more companies.’
(25)
a.
Mittu-izyoo-no
daigakui-ni
three-or.more-GEN university-DAT
sotugyoosei-ga syutugansita to]
graduate-NOM applied
that
Ken-ga Aya-ni
Ken-NOM Aya-DAT
itta.
told
[sokoi-no
it-GEN
b.
Mittu-izyoo-no
kaisyai-o
three-or.more-GEN company-ACC
syain-ga
tyoosasita
to]
employee-NOM investigated that
Ken-ga Aya-ni
Ken-NOM Aya-DAT
itta.
told
[sokoi-no
it-GEN
In (24) the pronominal is contained in the embedded subject and the QP is an embedded
object. In (25) the embedded object QP has undergone long-distance scrambling out of a
finite clause. The pronominal cannot be bound by the QP in (24) but can be in (25). Since
we know that scrambling out of a finite clause does not affect binding (see (4), (6), and
(19)), what makes variable binding possible in (25) must be an intermediate step of
long-distance scrambling, that is, scrambling within the embedded clause, on a par with step
(I) in (22)/(23). Thus, the claim that an intermediate step of long-distance scrambling within
an embedded clause can produce new binding relations is supported on independent
grounds.
Note that on this analysis, scrambling within the control clause never puts the QP in a
position c-commanding the pronominal in the case of (14) and (18), where the pronominal
is not contained in the controller. As a result, variable binding is impossible in those cases.
Therefore, this proposal argues strongly for the movement theory of control since there
would be no difference relevant to binding between the derivations of (8)/(10)/(11) and
those of (14)/(18) under a nonmovement approach to control, according to which the
controller is base-generated in the matrix clause and the subject of the control clause is an
independent element (i.e., PRO).
-86-
Scrambling and the Nature of Movement (Y. Takano)
4.
New Questions about the Nature of Movement
The proposal made in the previous section raises new questions for theoretical domains
related to movement. In this section, I will discuss four points bearing directly on the
present proposal. The purpose of this section is not to attempt to resolve them but to clarify
important issues for future research.
4.1.
On the Nature of Scrambling
One important claim in the present proposal is that scrambling out of a control clause
can never license variable binding. Why should this be the case? Given the claim in (21a),
this question boils down to the question why scrambling out of a clause cannot license
variable binding. The proposed analysis of scrambling out of a control clause has an
important consequence for this well-known issue.
One possible answer to the question why scrambling out of a clause cannot license
variable binding is (26).
(26)
Scrambling out of a clause is necessarily A’-movement.
This hypothesis has in fact been quite influential and entertained by many researchers (see
in particular Mahajan 1990 and Miyagawa 2005, 2006). One strong motivation for this
hypothesis comes from a contrast like that in (27).
(27)
a. * Whoi did hisi mother call t?
b.
Every boyi seems to hisi mother [t to be smart].
The example in (27a) shows a weak crossover effect. The point relevant here is that his
cannot be interpreted as a variable bound by who even though overt movement has put who
in a position that c-commands his. In contrast, his can be interpreted as bound by every boy
in (27b) due to overt movement of every boy (compare (27b) with *It seems to hisi mother
that every boyi is smart, which does not permit a bound variable interpretation for his).
The difference between (27a) and (27b) is usually attributed to the nature of movement:
whereas A-movement can license variable binding, A’-movement cannot. Along the same
lines, scrambling out of a clause cannot license variable binding if such scrambling is
necessarily A’-movement.
Note that approaches relying on (26) assume crucially that whereas clause-internal
scrambling can be A-movement (as in (2)), long-distance scrambling is necessarily
A’-movement. 9 Those approaches typically attribute the differential properties of
9
Cases like (i) below show that clause-internal scrambling can also be A’-movement. (Continued
on next page)
-87-
Nanzan Linguistics 5: Research Results and Activities 2008 ~ 2009
clause-internal and long-distance scrambling to the properties of their landing sites; namely,
clause-internal scrambling can target a specifier position, but long-distance scrambling must
target an adjoined position. The scrambled phrase can thus be in an A-position in the former
but in an A’-position in the latter. Given that this view claims that there are two different
types of scrambling, one involving movement to a Spec and another involving adjunction,
we might call it a nonuniform theory of scrambling.
Given the present proposal, the nonuniform theory cannot be the answer to the
question why scrambling out of a clause does not affect binding. Recall that the analysis of
(8)/(10)/(11) presented above relies crucially on the movement theory of control. More
specifically, movement of X in (22) and (23) plays a crucial role in making binding inside
the embedded clause possible. Note that here X moves from the embedded clause to a
θ-position in the matrix clause. Given that θ-positions are considered to be typical
A-positions, this is A-movement. This means that A-movement out of a control clause is
possible. Therefore, (26) cannot be correct for scrambling out of a control clause. If so, we
cannot appeal to a nonuniform theory of scrambling to account for why scrambling out of a
control clause does not license variable binding.
An alternative to a nonuniform theory is a uniform theory of scrambling, which treats
clause-internal and long-distance scrambling in the same way with respect to their landing
sites and attempts to account for differences between the two without appealing to (26). One
such theory has recently been proposed by Saito (2003, 2005) (other proposals for a
uniform theory of scrambling include those of Abe (1993), Bošković and Takahashi (1998),
Saito (1992), and Tada (1990, 1993)).
Saito (2003, 2005) proposes a theory of scrambling based on the idea that movement
chains are interpreted cyclically by means of deletion of features. Modifying and extending
Chomsky’s (1995: chap. 3) proposal for the formation of operator-variable structures by
deletion of parts of chains, Saito proposes that chains are interpreted as they are formed and
that chain interpretation deletes from a position of a chain all features that are not selected
in that position of the chain, where selection includes feature checking (agreement) and
θ-marking.
Let us consider a concrete case. Under the copy theory of movement, the sentence in
(28a) has the structure in (28b).
(i)
Zibunzisini-o Keni-ga
semeta.
self-ACC
Ken- NOM blamed
‘Ken blamed himself.’
If clause-internal scrambling were always A-movement, (i) would be a violation of condition (C),
contrary to fact.
-88-
Scrambling and the Nature of Movement (Y. Takano)
(28)
a.
Who did John see?
b.
[CP who
did John see who]
{P, O, A}
{P, O, A}
c.
[CP who
{P, O}
did John see who]
{A}
Assuming that each syntactic object is a set of features, Saito claims that the wh-phrase who
is a set of (at least) P(honological)-features, an O(perator)-feature, and an A(rgument)feature, which is closely tied with the referential properties of the phrase and participates in
binding relations.10 When who undergoes wh-movement, this feature set is copied, forming
a chain, so that there are two identical feature sets, one in Spec,C and another in the object
of the verb. Now chain interpretation applies. Since this is overt movement, the P-features
must be retained in the head of the chain and must be deleted in the tail (this is an essential
part of the definition of overt movement). Deletion of the rest of the features is contingent
on selection. The O-feature is selected in Spec,C (it enters into checking/agreement with C),
but is not selected in the object position, so that it is retained in the former and deleted in the
latter. By contrast, the A-feature is selected in the object position (the object is θ-marked
there), but is not selected in Spec,C. Thus, it is deleted in Spec,C. This results in the
structure in (28c), where the P- and O-features are located in Spec,C, whereas the A-feature
is located in the object position. Saito claims that the copy of who in Spec,C, having an
O-feature, functions as an operator and that that in the object, having an A-feature,
functions as a variable.
Long-distance wh-movement is analyzed in the same way. Consider (29).
(29)
a.
Who do you think John saw?
b.
[CP who
John saw who]
{P, O, A}
{P, O, A}
c.
[CP who
John saw who]
{P, O}
{A}
d.
[CP who do you think [CP who John saw who]]
{P, O}
{P, O}
{A}
e.
[CP who do you think [CP John saw who]]
{P, O}
{A}
Saito assumes that the sentence in (29a) is derived by two successive wh-movements, the
10
Saito calls the third feature a D-feature in Saito 2003 but an argument-feature in Saito 2005. I
follow the latter here.
-89-
Nanzan Linguistics 5: Research Results and Activities 2008 ~ 2009
first wh-movement being to Spec,C of the embedded clause and the second to Spec,C of the
matrix clause. The first wh-movement derives (29b), to which chain interpretation applies.
Here Saito follows Chomsky (2000) in assuming that this step of wh-movement takes place
because a feature of C attracts an O-feature of the wh-phrase. On this view, the O-feature is
selected in Spec,C. Therefore, the O-feature of who in the object position deletes. Deletion
of the P- and A-features takes place in the same way as clause-internal wh-movement.
Hence (29c) results. Then the second wh-movement applies, copying the feature set of who
in Spec,C of the embedded clause and deriving (29d). Chain interpretation applies to this
newly created chain. The P- and O-features are retained in Spec,C of the matrix clause for
the reasons stated above.11
For A-movement, consider (30).
(30)
Every boyi seems to hisi mother [t to be smart].
Saito assumes that A-movement is driven by an EPP-feature of T and that an EPP-feature
selects an A-feature. On these assumptions, (30) is derived as in (31).
(31)
a.
[TP every boy seems to his mother [TP every boy to be smart]]
{P, A}
{P, A}
b.
[TP every boy seems to his mother [TP every boy to be smart]]
{P, A}
{A}
Since the A-feature is selected in both the head and the tail of the A-movement chain, it is
retained in both positions. The raised DP every boy, retaining its A-feature, can bind his in
the matrix clause after A-movement. The result ensures that A-movement can license
variable binding.12
Regarding Japanese scrambling, Saito proposes that, regardless of whether it is
clause-internal or long-distance, it is a uniform operation targeting a specifier position (such
as Spec,T and Spec,C). He also assumes that it takes place without selection (that is, it is not
triggered by checking/agreement). This means two things. First, unlike wh-movement, there
is no O-feature involved in scrambling. Second, given that chain interpretation deletes the
A-feature from the head of the chain formed by scrambling, further scrambling copies only
11
Saito (2003) assumes with Chomsky (2000) that the feature of the intermediate C that attracted
the O-feature at step (29b) deletes after its selectional requirement is satisfied. As a result, the
O-feature is not selected in the intermediate Spec,C at step (29d) and hence it is deleted there in
(29e).
12
In (31) the DP every boy is assumed to move from Spec,T of the embedded clause. If it originates
in a lower position (such as the subject position of the small clause headed by smart), raising will
consist of two steps, the first step being to Spec,T of the embedded clause and the second step to
Spec,T of the matrix clause. Under Saito’s analysis, both steps of A-movement will be driven by an
EPP-feature of T.
-90-
Scrambling and the Nature of Movement (Y. Takano)
the P-features of the scrambled element.
Let us consider clause-internal scrambling first:
(32)
a.
Sono daigaku-ni
Ken-ga syutugansita.
that university-DAT Ken-NOM applied
‘Ken applied to that university.’
b.
[TP sono daigaku-ni Ken-ga sono daigaku-ni V]
{P, A}
{P, A}
c.
[TP sono daigaku-ni Ken-ga sono daigaku-ni V]
{P}
{A}
Since scrambling does not involve an operator feature, the scrambled element has no
operator feature and has only P- and A-features. They are copied when scrambling takes
place, giving rise to (32b). Since the A-feature is selected in the tail of the chain but not in
its head, it is deleted in the head, resulting in (32c).
Consider next long-distance scrambling out of a finite clause:
(33)
a.
Sono daigaku-ni
Masao-ga
Yumi-ni
that university-DAT Masao-NOM Yumi-DAT
[Ken-ga
syutugansita to] itta.
Ken-NOM applied
that told
‘Masao told Yumi that Ken applied to that university.’
b.
[CP sono daigaku-ni Ken-ga sono daigaku-ni V to]
{P, A}
{P, A}
c.
[CP sono daigaku-ni Ken-ga sono daigaku-ni V to]
{P}
{A}
d.
[TP sono daigaku-ni Masao-ga Yumi-ni
{P}
[CP sono daigaku-ni Ken-ga sono daigaku-ni V to]
{P}
{A}
e.
[TP sono daigaku-ni Masao-ga Yumi-ni
{P}
[CP Ken-ga sono daigaku-ni V to] V]
{A}
The derivation up to (33c) is identical to the derivation in (32), except that the landing site is
Spec,C in (33). Crucially, on this analysis, the element that has undergone clause-internal
-91-
Nanzan Linguistics 5: Research Results and Activities 2008 ~ 2009
scrambling has only P-features when it undergoes further scrambling and so scrambling out
of a clause moves (copies) only the P-features of the scrambled element. This is shown in
(33d). Given that the P-features are retained only in the head of the chain, those in the
intermediate Spec,C are deleted. The result is (33e).
As Saito argues, this analysis has the important consequence of deriving the effects of
“total reconstruction” induced by long-distance scrambling, examples of which are shown
in (4) and (6). As can be seen in (33d), scrambling out of a finite clause affects only the
P-features of the scrambled element (because the A-feature is deleted from the head of the
chain formed by clause-internal scrambling, as shown in (33b-c). Given that binding
requires the antecedent to have an A-feature, total reconstruction effects follow as a natural
consequence of this analysis, without appeal to any covert operations like reconstruction.13
At the same time, the very existence of the derivational point illustrated in (32b) ensures,
under a derivational approach to binding, that clause-internal scrambling can make binding
possible since the scrambled phrase has an A-feature right after clause-internal scrambling
(even though this A-feature gets deleted eventually). Thus, Saito’s uniform theory of
scrambling can account for the difference between clause-internal and long-distance
scrambling with respect to binding effects without appealing to (26).14,15
13
Shigeru Miyagawa (personal communication) pointed out that Saito’s analysis faces a problem in
dealing with anti-reconstruction effects as shown in (i) below.
(i)
Ken-ga
sotugyoosita daigaku-ni
kare-ga [Yumi-ga
syutugansuru to] omotteiru.
Ken-NOM graduated
university-DAT he-NOM Yumi-NOM apply
that think
‘He thinks that Yumi will apply to the university Ken graduated from.’
In this example the matrix subject pronoun kare can be coreferential with Ken, which is contained in
the phrase that has undergone long-distance scrambling. The problem is that if the phrase that has
undergone long-distance scrambling consists of only its P-features, Ken’s A-feature will stay in the
embedded clause and hence will be c-commanded by kare. If that is the case, coreference between
kare and Ken should violate condition C. I suspect that the problem can be resolved in the following
way. First, Saito (2005: note 3) suggests that each syntactic object has a categorial feature and that
categorial features are represented at every position of a chain. Assuming now an approach that
accounts for anti-reconstruction effects by allowing relative clauses to be late-merged with moved
elements (Lebeaux 1988, Nishigauchi 2002, Miyagawa 2006), it is possible to analyze (i) as
involving late-merger of the relative clause containing Ken with the scrambled DP, which retains its
categorial feature. This derivation will ensure that (i) does not violate condition C. I also suggest
that the relative clause can be construed with the relative head daigaku ‘university,’ whose A-feature
is retained at the tail of the chain, because the relative clause and the relative head are contained in
the same chain.
14
To deal with cases like (i) of note 9, repeated below, Saito (2003) claims that, unlike condition
(A), which is an anywhere condition and so is satisfied derivationally, condition (C) is an LF
condition, so that it applies after chain interpretation. (Continued on next page)
-92-
Scrambling and the Nature of Movement (Y. Takano)
Under Saito’s theory, scrambling out of a control clause will be analyzed in the same
way as scrambling out of a finite clause illustrated in (33). If so, it follows straightforwardly
that scrambling out of a control clause does not license variable binding since this
scrambling will involve movement of the P-features alone. Thus, Saito’s uniform theory is a
promising direction to pursue to derive the relevant properties of scrambling out of a control
clause.
At the same time, however, details need to be worked out to make Saito’s theory
compatible with a general theory of successive-cyclic movement, given the results of much
recent work on successive-cyclic movement (see Bošković 2007 and references cited there
for various issues relevant to successive-cyclic movement). For instance, Bošković (2007)
argues against the idea that successive-cyclic movement is driven by a feature of the head to
whose specifier movement takes place. He proposes instead that all successive-cyclic
movement is driven by an uninterpretable feature of the moving element. Moreover, he
suggests that movement proceeds by way of a specifier of each intermediate head (in other
words, every maximal projection is a phase/barrier). These aspects of successive-cyclic
movement, if correct, are not compatible with Saito’s theory of movement in its present
form. Recall that in Saito’s theory successive-cyclic movement, whether wh-movement or
A-movement, is triggered by a feature of the head to whose specifier movement takes place.
This is in conflict with Bošković’s claim that successive-cyclic movement is always driven
by a feature of the moving element. Another question arises about the timing of chain
interpretation: when exactly does chain interpretation (in terms of deletion) apply if
movement always proceeds by way of each intermediate specifier, as Bošković suggests?
Thus, if the theory of successive-cyclic movement proposed by Bošković is on the right
track and if we are to maintain Saito’s uniform theory of scrambling to explain the
properties of Japanese scrambling discussed here, then we need to seek a way to make them
compatible with each other.
(i)
Zibunzisini-o Keni-ga
semeta.
self-ACC
Ken- NOM blamed
‘Ken blamed himself.’
(ii) Zibunzisini-o Keni-ga zibunzisin-o semeta.
{P}
{A}
As shown in (ii), after chain interpretation, zibunzisin ‘self’ in the chain head does not have its
A-feature and hence does not bind Ken, as a result of which a condition (C) violation does not occur.
15
Note also that Saito’s analysis has a consequence for the issue mentioned in note 8. On this
analysis, A-movement of the controller past the scrambled phrase is possible because the scrambled
phrase, having had its A-feature deleted, has lost its argument status when A-movement of the
controller takes place and hence is invisible to that movement.
-93-
Nanzan Linguistics 5: Research Results and Activities 2008 ~ 2009
4.2.
Illicit Scrambling
Recall that the present analysis of (8)/(10)/(11) rests crucially on the interaction of
(clause-internal) scrambling and movement of the controller under the movement theory of
control. The relevant part of the analysis is repeated in (34).
(34)
(II) movement of controller
X [a Y
<X> <Y> V]
V
(I) scrambling
I proposed that clause-internal scrambling of the embedded object Y makes it possible for Y
to bind a pronominal contained in the controller X, which moves from within the embedded
clause to the matrix clause. In the relevant examples Y undergoes further scrambling out of
the embedded clause, so that it ends up appearing in front of X in the matrix clause. A
question arises here.16 What happens if Y does not scramble further out of the embedded
clause? The resulting sentence would have the word order X-Y-V-V, which corresponds
to the order in (7a), for instance, repeated below.
(7)
a. * Sokoi-no sotugyoosei-ga [ mittu-izyoo-no
daigakui-ni
it-GEN
graduate-NOM
three-or.more-GEN university-DAT
syutugansi-yoo to] sita.
apply-will
that did
‘Their graduates tried to apply to three or more universities.’
When analyzing (7a), we always assume that it does not involve scrambling of the
embedded object. In that way, we can capture the ungrammaticality of this example. But if
the derivation in (34) (with no further scrambling of Y) were available, (7a) would be as
acceptable as (8), repeated below, contrary to fact.
(8)
a.
Mittu-izyoo-no
daigakui-ni
sokoi-no sotugyoosei-ga
three-or.more-GEN university-DAT it-GEN
graduate-NOM
[syutugansi-yoo to] sita.
apply-will
that did
We thus need to exclude this derivation.
In fact, the problem is more general. Problems of the same nature arise independently
of the analysis of the control cases in question. Consider the following derivation for a
simplex sentence:
16
Kensuke Takita (personal communication) first brought this question to my attention.
-94-
Scrambling and the Nature of Movement (Y. Takano)
(35)
a.
[a Y X <Y> V]
b.
[ X Y <X> <Y> V]
In (35a) the object Y scrambles over the subject X and in (35b) the subject X scrambles
over the scrambled Y. This derivation will result in the word order X-Y-V, which is
identical to an SOV sentence without scrambling. If this derivation were possible, cases like
(1a), repeated below, would be acceptable (Y could bind into X due to the existence of the
step in (35a)).
(1)
a. * Sokoi-no sotugyoosei-ga mittu-izyoo-no
daigakui-ni
syutugansita.
it-GEN
graduate-NOM three-or.more-GEN university-DAT applied
‘Their gradutaes applied to three or more universities.’
Derivations like that in (35) are a long-standing problem for any analysis of Japanese
scrambling and there have been a number of proposals made to deal with this problem.
For example, Saito (1985) proposed that the step in (35b) is disallowed because nominative
phrases cannot scramble in Japanese. If so, the problem for (1a) will not arise. However,
this account does not cover the problem with the derivation in (34), where the movement of
X is guaranteed by the movement theory of control.
Another approach to the problem in question was suggested by Hoji (1985). He put
forth the following condition:
(36)
A syntactic adjunction operation cannot apply if it does not change the order of the
overt lexical string. (Hoji 1985: 352)
Assuming that scrambling is an adjunction operation, Hoji claims that this condition blocks
the applications of scrambling in (35) since these applications of scrambling do not change
the original word order of the subject and the object.17 This approach can be extended to
cover the problem with (34) if we interpret Hoji’s condition as stating that if the surface
form of a given sentence corresponds to a string that can be analyzed without scrambling,
the sentence is indeed understood to involve no scrambling. Since the sentence in (7a) can
be analyzed without scrambling, the derivation in (34) with “superfluous scrambling” is
blocked
Recently, Ko (2007) proposes a different line of analysis that excludes derivations like
17
Hoji (1985: 367) suggests that the condition in (36) may fall outside formal grammar and belong
to a domain of parsing. Abe (1993) and Takano (1992) propose to derive Hoji’s condition from
economy of derivation. See also Takano 2007 for a different approach capitalizing on the properties
of the optional assignment of a feature that triggers scrambling.
-95-
Nanzan Linguistics 5: Research Results and Activities 2008 ~ 2009
(34) and (35). On her proposal, the derivation in (35) is blocked by Fox and Pesetsky’s
(2003) Linearization Preservation, given in (37), where Spell-Out Domain is a syntactic
constituent relevant to the determination of linear order of syntactic elements.18
(37)
The linear ordering of syntactic units is affected by Merge and Move within a
Spell-out Domain, but is fixed once and for all at the end of each Spell-out Domain.
The basic idea here is that linear order is determined cyclically and that linear order
established at the end of a given Spell-out Domain must be preserved at the end of each
later Spell-out Domain. Let us imagine that a in (35a) is a Spell-out Domain.19 At the end
of this Spell-out Domain, the order between X and Y is fixed in such a way that Y precedes
X. By Linearization Preservation in (37), this order between X and Y must be preserved at
later Spell-out Domains. Therefore, the derivation in (35b) is disallowed since it changes
the order between X and Y fixed at the earlier Spell-out Domain. Thus, the sentence in (1b)
cannot have the derivation in (35).20 This approach can account for the illicit derivation in
18
Ko’s (2007) proposal is intended to account for facts related to quantifier float in Korean, but can
readily be extended to the cases at hand. See also Takita 2008 for a proposal that Linearization
Preservation can account for scope effects with scrambling in Japanese.
19
Ko (2007) claims that vP and CP are Spell-out Domains. The exact identity of Spell-out Domains
does not concern us here.
20
Consider the following problem, which was originally discussed by Takita (2008) in the context
of scope interactions of QPs in Japanese:
(i)
a.
[a X Y V]
b.
Y [a X <Y> V]
c.
X Y [a <X> <Y> V]
At the end of the Spell-out Domain a, X precedes Y. Then Y scrambles out of a, followed by
scrambling of X. The resulting order between X and Y in (ic) preserves their order determined at a
in (ia). So this derivation satisfies Linearization Preservation. If the step in (ib) guaranteed Y’s
binding into X, (1a) would allow variable binding with this derivation. The fact that it does not
suggests that Y can bind into X only within a. We can imagine a number of ways to ensure this.
One possibility is to appeal to Saito’s (2003, 2005) theory of scrambling, according to which
scrambling out of a certain domain has no binding effects (see section 3.1). If a is the relevant
domain, scrambling out of a will have no binding effects and so Y will bind into X only within a.
Two other possibilities were suggested to me by Mamoru Saito. Suppose a is a domain relevant to
interpretation (as well as a Spell-out Domain). Following Chomsky (2000, 2001), let us call such a
domain a phase. We might then claim that binding relations are established only at the end of each
phase. On this view, Y does not bind into X in the derivation in (i) because Y does not c-command
X at the phase a in (ia) or at the next higher phase in (ic) (here we assume a derivational approach to
binding in which only chain heads enter into binding, so that the copies of X and Y inside a in (ic)
are irrelevant). Alternatively, we might entertain the hypothesis that binding relations established at
a given phase must be preserved throughout a derivation. On this approach, Y does not bind into X
-96-
Scrambling and the Nature of Movement (Y. Takano)
(34) as well. Suppose that a in (34) is a Spell-out Domain. Then the order between X and Y
is fixed there, that is, Y precedes X. But this order is not preserved at a later Spell-out
Domain if X moves over Y and Y does not move, as in (34). Thus, the derivation in (34)
violates Linearization Preservation and hence (7a) cannot be derived in this way.21
As we have seen, both the approach invoking Hoji’s (1985) condition in (36) and the
approach appealing to Linearization Preservation can exclude the undesired derivations in
(34) and (35). The two approaches are equal on this count. However, we can provide
empirical evidence for the second approach based on binding effects with scrambling out of
a control clause.
Observe first the following case (NC = nominal complementizer):
(38) ? Mittu-izyoo-no
daigakui-ni
sokoi-no sotugyoosei-ga kotosi
three-or.more-GEN university-DAT it-GEN graduate-NOM this.year
[ rainen
syutugansuru koto]-o kessinsita.
next.year apply
NC-ACC decided
‘Their graduates decided this year to apply to three or more universities next year.’
This case involves scrambling of the embedded object out of a control clause, which makes
it possible for the scrambled object to bind into the matrix subject, which is a controller.
On the present analysis, the acceptability of variable binding indicates that the embedded
object first scrambled within the embedded clause. Now observe the case in (39).22
(39)?*Sokoi-no sotugyoosei-ga
it-GEN graduate-NOM
[ rainen
syutugansuru
next.year apply
mittu-izyoo-no
daigakui-ni
kotosi
three-or.more-GEN university-DAT this.year
koto]-o kessinsita.
NC-ACC decided
Here the embedded object scrambles out of the control clause to a position between the
matrix subject and the matrix adjunct. Unlike the example in (38), this example does not
permit a bound variable interpretation for the pronominal contained in the matrix subject
(though scrambling of the embedded object itself is fine, as evidenced by the fact that (39)
is grammatical if the matrix subject is replaced by Ken).
The fact that (39) does not allow a bound variable interpretation poses a serious
problem for the approach that blocks the derivation in (34) by appealing to (36). Notice that
in the case of (39), the embedded object scrambles to the matrix clause, as is clearly
at phase a in (ia) and this binding relation must be preserved throughout the derivation. As a result,
scrambling of Y out of a has no effects on binding.
21
The same reservation applies here as the previous note.
22
I am indebted to Norvin Richards for this example.
-97-
Nanzan Linguistics 5: Research Results and Activities 2008 ~ 2009
indicated by its position in front of the matrix adjunct. Since this scrambling of the
embedded object does affect word order, from the perspective of (36), it should be allowed
with the intermediate step in (34). But then the impossibility of the bound variable reading
cannot be accounted for.
In contrast, the derivation in (34) is correctly excluded for (39) under the approach
invoking Linearization Preservation. On this approach, the order between X and Y
determined at a in (34), namely, the embedded object preceding the controller, must be
preserved. But in (39) the controller precedes the embedded object, in violation of
Linearization Preservation. Therefore, (39) cannot involve the derivation in (34).
These considerations show that scrambling out of a control clause provides
independent empirical evidence for an approach like Ko’s (2007) that constrains the
application of scrambling with Linearization Preservation.
On the other hand, an important conceptual question arises with Linearization
Preservation. Juan Uriagereka pointed out to me that reliance on Linearization Preservation
presupposes that language has “counters,” which is not a desirable move, given the fairly
common assumption to the contrary. If so, we are in a dilemma: facts about Japanese
scrambling do appear to support Linearization Preservation, but the latter seems to be
problematic on conceptual grounds. Therefore, it is very important for future work to
consider whether the facts discussed in this subsection indeed can only be explained by
Linearization Preservation and if yes, what exactly the status of Linearization Preservation
is in the theory of human language.
4.3.
Long-Distance Scrambling: Control vs. Finite Clauses
Recall that the analysis proposed in section 3 relies heavily on the claim in (21a),
repeated below.
(21)
a.
Scrambling out of a control clause patterns with scrambling out of a finite
clause.
As we have seen, facts involving variable binding support this claim. But there are cases
where scrambling out of a control clause does pattern differently from scrambling out of a
finite clause. One such case is so-called “additional-wh effects” (see Saito 1994 for
extensive discussion and an analysis of additional wh-effects in Japanese).
To see this, let us first consider (40).
(40)
a. * Naze dare-ga
sono hon-o
katta no.
why who-NOM that book-ACC bought Q
‘Who bought the book why?’
-98-
Scrambling and the Nature of Movement (Y. Takano)
b.
Dare-ga naze sono hon-o
katta no.
who-NOM why that book-ACC bought Q
c.
Nani-o
naze dare-ga
katta no.
what-ACC why who-NOM bought Q
‘Who bought what why?’
The contrast between (40a) and (40b) shows that the Japanese wh-phrase naze cannot
c-command an argument wh-phrase. Moreover, the grammaticality of (40c) shows that this
effect can be voided by the presence of an additional argument wh-phrase c-commanding
naze. Note that (40c) also indicates that a scrambled phrase can serve as an additional
wh-phrase.
Not all scrambled phrases can serve as additional wh-phrases, however. Consider (41).
(41)
a. * Naze dare-ga
Ken-ni [ Yumi-ga
sono hon-o
katta to] itta no.
why who-NOM Ken-DAT Yumi-NOM that book-ACC bought that said Q
‘Why did who tell Ken that Yumi bought that book?’
b. *Nani-o naze dare-ga Ken-ni [ Yumi-ga katta to] itta no.
what-ACC why who-NOM Ken-DAT Yumi-NOM bought that said Q
(41a) is ungrammatical because naze c-commands dare. In (41b) the object of the
embedded clause is a wh-phrase and is scrambled to the front of the matrix clause. The
sentence is still ungrammatical even though the scrambled phrase c-commands naze. This
fact indicates that long-distance scrambled phrases cannot serve as additional wh-phrases.
In the case of (41b), the embedded object has scrambled out of a finite clause.
Nemoto (1993) observes that a different patter emerges when the wh-phrase scrambles out
of a control clause.
(42)
a. * Naze dare-ga
Ken-ni [ sono hon-o
kau yoo(ni)] susumeta
no.
why who-NOM Ken-DAT that book-ACC buy C
recommended Q
‘Why did who recommend Ken to buy that book?’
b.
Nani-o
naze dare-ga
Ken-ni [ kau
what-ACC why who-NOM Ken-DAT buy
yoo(ni)] susumeta
no.
C
recommended Q
The sentence in (42b), where the wh-phrase has scrambled out of a control clause, does
improve on the sentence in (42a). Here we see a contrast between two types of
long-distance scrambling: scrambling out of a finite clause does not induce additional
wh-effects, but scrambling out of a control clause does.
This is unexpected under the view in (21a). Recall that this view is supported by
-99-
Nanzan Linguistics 5: Research Results and Activities 2008 ~ 2009
binding facts. A detailed investigation is thus necessary to find out why binding and
additional wh-effects should differ in this way.
4.4.
Bound Pronoun Effects
As a final point, let us discuss a difference between Japanese scrambling and English
wh-movement. Let us first consider the following well-known fact:
(43)
Who did everyone buy for Max?
May (1985) observes that the sentence in (43) permits a reading on which the value of who
covaries with that of everyone. Thus, (43) can be interpreted as asking for a list of pairs that
specifies which person bought what. Lasnik (2007) calls this a “family of questions (FOQ)
reading.”
The FOQ reading is also possible with long-distance wh-movement, as in (44).
(44)
Who do you think everyone saw?
However, May (1977) and Sloan (1991) observe that the FOQ reading is not available with
long-distance wh-movement if the QP everyone is placed in the subject of the matrix clause,
as in (45).
(45)
Who does everyone think Mary saw?
Thus, long-distance wh-movement does not give rise to an FOQ reading if a QP is in the
subject of the matrix clause.
Sloan (1991) points out an interesting exception to this generalization (Sloan attributes
the observation to Robert May). The example in (46), though it involves long-distance
wh-movement and a QP in the matrix subject, does allow an FOQ reading, in contrast to
(45).
(46)
Who does everyonei think hei saw?
The important property that distinguishes (46) from (45) is the fact that the subject of the
embedded clause is a pronoun bound by the QP in the matrix subject. In fact, the FOQ
reading is available in (46) only on the reading on which the embedded subject is bound by
the matrix subject. Thus, the presence of a bound pronoun in the embedded subject makes
possible the FOQ reading with long distance wh-movement that is otherwise unavailable.23
This situation with English wh-movement contrasts with what we saw with Japanese
23
Lasnik (2007) observes that similar effects induced by bound pronouns can be found with
quantifier scope, gapping, antecedent-contained deletion, reciprocal binding, extraposition, and
multiple sluicing.
-100-
Scrambling and the Nature of Movement (Y. Takano)
scrambling. Recall that in (19), repeated below, long-distance scrambling does not make
variable binding possible even if pro in the embedded subject is bound by the matrix
subject.
(19)
a. ?* Mittu-izyoo-no
three-or.more-GEN
[proj syutugansita
applied
daigakui-ni
sokoi-no sotugyooseij-ga Ken-ni
university-DAT it-GEN
graduate-NOM Ken-DAT
to] itta.
that said
‘Their graduates told Ken that they applied to three or more universities.’
b.?* Mittu-izyoo-no
three-or.more-GEN
[proj tyoosasita
investigated
kaisyai-o
sokoi-no syainj-ga
Ken-ni
company-ACC it-GEN
employee-NOM Ken-DAT
to] itta.
that said
‘Their employees told Ken that they investigated three or more companies.’
Why does Japanese scrambling differ from English wh-movement in this respect? This is
another important issue that needs to be addressed and resolved in future work.
5.
Summary
On the basis of the discovery of new facts about binding effects with scrambling out of
a control clause in Japanese, I have proposed an analysis of those facts consisting of two
major claims:
(47)
a.
Scrambling out of a control clause patterns with scrambling out of a finite clause
with respect to binding effects.
b.
Obligatory control involves movement of the controller.
The present study thus provides a new argument in their favor. I have also pointed out that
this proposal raises new questions about the nature of movement that need to be resolved in
future study.
References
Abe, J. (1993) Binding Conditions and Scrambling without A/A’ Distinction, Ph.D. dissertation,
University of Connecticut.
Boeckx, C. (2000) “A Note on Contraction,” Linguistic Inquiry 31, 357-366.
Boeckx, C. and N. Hornstein (2003) “Reply to Control Is Not Movement,” Linguistic Inquiry 34,
269-280.
Boeckx, C. and N. Hornstein (2004) “Movement under Control,” Linguistic Inquiry 35, 431-452.
-101-
Nanzan Linguistics 5: Research Results and Activities 2008 ~ 2009
Boeckx, C. and N. Hornstein (2006) “The Virtues of Control as Movement,” Syntax 9, 118-130.
Bošković, Ž. (2004) “Topicalization, Focalization, Lexical Insertion, and Scrambling,” Linguistic
Inquiry 35, 613-638.
Bošković, Ž. (2007) “On the Locality and Motivation of Move and Agree: An Even More Minimal
Theory,” Linguistic Inquiry 38, 589-644.
Bošković, Ž. and D. Takahashi (1998) “Scrambling and Last Resort,” Linguistic Inquiry 29,
347-366.
Bowers, J. (1973) Grammatical Relations, Ph.D. dissertation, MIT.
Bowers, J. (2008) “On Reducing Control to Movement,” Syntax 11, 125-143.
Chomsky, N. (1995) The Minimalist Program, MIT Press, Cambridge, Mass.
Chomsky, N. (2000) “Minimalist Inquiries: The Framework,” in R. Martin, D. Michaels, and J.
Uriagereka, eds., Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik,
MIT Press, Cambridge, Mass, 89-155.
Chomsky, N. (2001) “Derivation by Phase,” in M. Kenstowicz, ed., Ken Hale: A Life in Language,
MIT Press, Cambridge, Mass, 1-52.
Fox, D. and D. Pesetsky (2003) “Cyclic Linearization and the Typology of Movement,” Lecture
Notes, Linguistic Society of America Summer Institute, Michigan State University.
Fukui, N. (1986) A Theory of Category Projection and Its Applications, Ph.D. dissertation, MIT.
Fukui, N. (1993) “Parameters and Optionality,” Linguistic Inquiry 24, 399-420.
Fujii, T. (2006) Some Theoretical Issues in Japanese Control, Ph.D. dissertation, University of
Maryland.
Grewendorf, G. and J. Sabel (1999) “Scrambling in German and Japanese: Adjunction versus
Multiple Specifiers,” Natural Language and Linguistic Theory 17, 1-65.
Hoji, H. (1985) Logical Form Constraints and Configurational Structures in Japanese, Ph.D.
dissertation, University of Washington.
Hoji, H. (2003) “Falsifiability and Repeatability in Generative Grammar: A Case Study of Anaphora
and Scope Dependency in Japanese,” Lingua 113, 377-446.
Hornstein, N. (1998) “Movement and chains,” Syntax 1, 99-127.
Hornstein, N. (1999) “Movement and Control,” Linguistic Inquiry 30, 69-96.
Hornstein, N. (2001) Move! A Minimalist Theory of Construal, Blackwell, Oxford.
Hornstein, N. (2003) “On Control,” in R. Hendrick, ed., Minimalist Syntax, Blackwell, Oxford,
6-81.
Kawamura, T. (2004) “A Feature-Checking Analysis of Japanese Scrambling,” Journal of
Linguistics 40, 45-68.
Kitahara, H. (2002) “Scrambling, Case, and Interpretability,” in S. D. Epstein and T. D. Seely, eds.,
Derivation and Explanation in the Minimalist Program, Blackwell, Malden, 167-183.
Ko, H. (2007) “Asymmetries in Scrambling and Cyclic Linearization,” Linguistic Inquiry 38, 49-84.
Kuroda, S.-Y. (1988) “Whether We Agree or Not: A Comparative Syntax of English and Japanese,”
Linguisticae Investigationes 12, 1-47.
Lasnik, H. (2007) “A Family of Questions,” unpublished ms., University of Maryland.
Mahajan, A. (1989) “On the A/A-bar Distinction: Scrambling and Weak Crossover in Hindi,”
unpublished ms., MIT.
Mahajan, A. (1990) The A/A-bar Distinction and Movement Theory, Ph.D. dissertation, MIT.
-102-
Scrambling and the Nature of Movement (Y. Takano)
May, R. (1977) The Grammar of Quantification, Ph.D. dissertation, MIT.
May, R. (1985) Logical Form: Its Structure and Derivation, MIT Press, Cambridge, Mass.
Miyagawa, S. (1997) “Against Optional Scrambling,” Linguistic Inquiry 28, 1-25.
Miyagawa, S. (2005) “EPP and Semantically Vacuous Scrambling,” in J. Sabel and M. Saito, eds.,
The Free Word Order Phenomenon: Its Syntactic Sources and Diversity, Mouton de
Gruyter, Berlin, 181-220.
Miyagawa, S. (2006) “On the “Undoing” Property of Scrambling: A Response to Bošković,”
Linguistic Inquiry 37, 607-624.
Nemoto, N. (1993) Chains and Case Positions: A Study from Scrambling in Japanese, Ph.D.
dissertation, University of Connecticut.
Nishigauchi, T. (2002) “Scrambling and Reconstruction at LF,” Gengo Kenkyu 121, 49-105.
Oka, T. (1989) “On the Spec of IP,” unpublished ms., MIT.
O’Neil, J. (1995) “Out of Control,” NELS 25, 361-371.
Saito, M. (1985) Some Asymmetries in Japanese and Their Theoretical Implications, Ph.D.
dissertation.
Saito, M. (1989) “Scrambling as Semantically Vacuous A’-movement,” in M. R. Baltin and A. S.
Kroch, eds., Alternative Conceptions of Phrase Structure, University of Chicago Press,
Chicago, 182-200.
Saito, M. (1992) “Long-distance Scrambling in Japanese,” Journal of East Asian Linguistics 1,
69-118.
Saito, M. (2003) “A Derivational Approach to the Interpretation of Scrambling Chains,” Lingua 113,
481-518.
Saito, M. (2005) “Further Notes on the Interpretation of Scrambling Chains,” in J. Sabel and M.
Saito, eds., The Free Word Order Phenomenon: Its Syntactic Sources and Diversity,
Mouton de Gruyter, Berlin, 335-376.
Saito, M. and N. Fukui (1998) “Order in Phrase Structure and Movement,” Linguistic Inquiry 29,
439-474.
Sauerland, U. (1999) “Erasability and Interpretation,” Syntax 2, 161-188.
Sloan, K. (1991) “Quantifier-Wh Interaction,” MIT Working Papers in Linguistics 15, 219-237.
Tada, H. (1990) “Scrambling(s),” unpublished ms., MIT.
Tada, H. (1993) A/A-Bar Partition in Derivation, Ph.D. dissertation, MIT.
Takano, Y. (1992) “Economy of Derivation: A View from Scrambling,” paper presented at Arizona
Linguistic Conference.
Takano, Y. (1995) “Scrambling, Relativized Minimality, and Economy of Derivation,” WCCFL 13,
385-399.
Takano, Y. (1998) “Object Shift and Scrambling,” Natural Language and Linguistic Theory 16,
817-889.
Takano, Y. (2007) “Making Rightward Scrambling Possible,” Kinjo Gakuin Daigaku Ronshu,
Studies in Humanities vol. 3, 17-58.
Takita, K. (2008) “String-Vacuous Scrambling and Cyclic Linearization,” Proceedings of the Ninth
Tokyo Conference on Psycholinguistics, Hituzi Syobo, Tokyo, 225-249.
Uchibori, A. (2000) The Syntax of Subjunctive Complements: Evidence from Japanese, Ph.D.
dissertation, University of Connecticut.
-103-
Nanzan Linguistics 5: Research Results and Activities 2008 ~ 2009
Yamashita, H. (2006) “A-Type Movement in Japanese and the EPP,” Proceedings of the 8th Seoul
International Conference on Generative Grammar (SICOGG): Minimalist Views on
Language Design, 333-352.
-104-